How to measure attention on social with data you already have

The attention data most brands say they lack is already in their ad account, because watch time is how Meta and TikTok decide what stays on screen, every session, and the same instrument scores every ad you run. Meta's own attention deck opens with a candid slide. Viewability is the opportunity to see an ad, it has a standard, the Media Rating Council's 50% of pixels for two continuous seconds, and it is measurable. Attention is whether the ad was actively perceived and watched, it has no industry standard, and it is hard to measure. Forty pages of commissioned vendor studies later the deck concludes that more evidence is needed, a strange modesty from the platform that sells the media, because the most practised attention measurers in the world are the platforms themselves.

The feed is an attention instrument

The ranking systems that assemble a feed predict, for each person and each piece of content, how long they will stay with it and what they will do next, and serve whatever maximises time in the session. They are trained on billions of watch decisions a day and corrected against what people actually did, and an ad entering that system is scored on the same instruments, which is why Meta's auction ranks ads on estimated action rate and quality as well as bid. An advertiser reading watch metrics is reading the platform's own attention instrument, at impression scale, at no extra cost, for every ad every day. The same deck concedes what the instrument does not do, average view time per impression does not represent how much people actually watched, because the distribution matters. That is an argument for reading the distribution, stop rate, hold rate and watched rate per ad as set out in How to read paid social results, not for ignoring the instrument.

Watching is a choice

The reason platform play metrics carry weight on this medium and did not on television is the skip. A rating point said a television was on in a room and nobody could know whether anyone was looking. On a feed, continuing to watch is a decision the viewer takes every second against the pull of the next thing, so a play that ran past three seconds and on to fifteen is evidence of a choice, repeated.

The standard makes the argument in probability terms. Platforms measure play, not eyes. But Lumen's cross-media panel, from a company that sells attention measurement, puts the population average at roughly two seconds of play for every attentive second on Meta and TikTok feeds, and that is an average across every impression including the ones scrolled past in half a second, so conditional on a three-second play the attention density is materially higher, and conditional on a fifteen-second play higher again. Dentsu's work with Kantar and Lumen, from companies that trade on attention metrics, found that once an ad is watched past two seconds the viewer has effectively chosen not to skip, and that voluntary attention works harder than forced attention. So the share of impressions that became watched views is a statistically sound claim about active attention, with a confidence a served impression never had. The Havas, Lumen and Brand Metrics analysis of 9,089 campaigns puts the boundary plainly, "there is a big difference between having an opportunity to see an ad and actually looking at that ad", and on a skippable medium the platform's own play data is the first instrument on the right side of that line. TikTok built the same logic into a metric, a focused view counts at six seconds, at completion, or on a positive interaction inside the first six seconds, a threshold below which the platform will not count the view as attentive and above which it bills for it.

Attention is not engagement

Likes, comments, shares and clicks are actions, a minority of people take them, and that minority skews to fans and heavy buyers who would have bought anyway, which is why Ehrenberg-Bass is right to dismiss engagement as a brand-building metric and why the standard files interaction rate as the textbook vanity metric. Watching is different in kind. It is the absence of an action, the person did not skip, and it is measured across every impression, including the light buyers who will never follow a brand or comment on anything and who are the audience growth depends on. A watched second with the distinctive assets on screen is a branded second, so watched rate across a portfolio is branded reach, and served reach was only the opportunity for it. The instrument counts the quiet majority.

Effective frequency

Where the instrument earns its keep first is frequency. A reported frequency of six sounds like a well-exposed audience. If each exposure ended in a two-second skip, the buyer experienced the brand for a combined twelve seconds in six fragments, and what those fragments can do is specific. A skip can remind. If the distinctive assets were on screen in the first second, each pass refreshes a memory that already exists, the refreshing job Byron Sharp describes for established brands, and it is real work. A skip cannot build. An association that is not there yet needs the watched view, the seconds past the threshold where new memory forms, which Amplified Intelligence, a company that sells attention measurement, puts at about 2.5 seconds of active attention. So the reported frequency is six, the reminding frequency may be six, the building frequency may well be zero, and the tracker is flat because, in the sense that matters to a brand trying to grow, the campaign barely reached that person.

Run the numbers on one ad. A thousand impressions at a 25% stop rate gives 250 plays past three seconds. A 40% hold rate takes 100 of those to a ThruPlay at fifteen seconds. Watched rate, 10%. Serve that ad six times to the same thousand people and the reported frequency is six, but the expected number of watched exposures per person is 0.6, and most people received none. Now raise the watched rate to 20% with a stronger opening and body, the same media plan delivers 1.2 watched exposures per person, and the number of people who received at least one has roughly doubled. Nothing in the media buy changed. The frequency conversation, where teams debate caps to avoid wearing out an audience most of whom never watched, has been pointed at the wrong number. Stop capping served frequency and start managing watched frequency.

Necessary, not sufficient

The instrument has a limit and the standard respects it. Attention is the condition under which an ad can work, not the ad working. The Advertising Research Foundation's 2026 validation study across four national campaigns found that media weight plays a critical role in translating attention into outcomes, that lower-attention channels sometimes delivered meaningful recall gains when they received sufficient weight and reach, and that attention should be read alongside media weight, reach and frequency rather than as a stand-alone proxy. Meta's Lumen data across 72 ads and 12 advertisers found that in three cases out of four the highest-attention format did not deliver the highest brand lift, and the Nielsen marketing mix meta-analysis Meta commissioned across 119 CPG brands found view time a weak predictor of sales impact next to frequency, placement and reach.

None of this argues against reading attention, it is the reason the standard reads results in order. Stop rate says the reminding job was done and the building job made possible, watched rate says the building job was made probable, and only a lift study says it happened. Why every social ad has two jobs sets out what each number reads. Performance is the leading indicator, read weekly, and effectiveness is the verdict, read on its own clock.

Eye-tracking purism

Attention is a property of a person, not of a video player, the purist says, and a play counter cannot tell whether eyes were on the screen, whether the viewer was reading a caption over the ad or looking at it at all. Only measured gaze, eye tracking in a lab or through a panel's front camera, is attention data, and everything else is a proxy that flatters whoever is selling it. Amplified Intelligence, Lumen, Realeyes and Playground XYZ, who all sell attention measurement, have built their businesses on that position and their data has taught the industry most of what it knows about attention and memory.

The position is right about what gaze measures and wrong about what follows. Eye tracking runs on panels of thousands and lab studies of hundreds, it is expensive, it is not available per ad per day, and by Meta's own summary its outputs depend on the lab set-up used. Play data is available at impression scale for every ad, and on a medium where playing on is a choice, its relationship to measured attention is what the attention vendors' own benchmarks establish. The right use of eye tracking is calibration, what a second of play is worth in attention on this platform and format, and the right use of play data is management, which ads earned that attention this week. Treating the calibration instrument as the only legitimate one leaves an account with no attention data on the days it needs to decide.

Social measures attention in every session, the platforms have been acting on it for a decade, and the instrument that tells you which of your ads is being watched is already in the account. Media creates the opportunity to see your ads. Creative determines whether they work.

What to do with this

Pull every ad from the last quarter, one platform, one threshold, Meta ThruPlays over impressions or TikTok focused views over impressions and never the two blended, and rank by watched rate. Rebuild the plan's headline numbers on a watched basis, watched reach instead of served reach, watched frequency instead of served frequency, cost per watched impression instead of CPM. The numbers shrink, sometimes dramatically, and the levers that move them, the opening, the body, the variety of the portfolio, are creative levers the old currency could not see. Check the proxy against the lift study whenever one exists, because the instrument predicts and only the study knows. How to audit your own paid social account turns this into a worksheet, and What the numbers show when creative is managed weekly is what we have seen the gap between best and worst ad do once a weekly loop is pointed at it.

Questions

Is a ThruPlay the same thing as attention?

No, and the standard does not call it that. A ThruPlay is fifteen seconds of play or completion on Meta, a focused view is six seconds or completion on TikTok, and both count video running in a viewport, not eyes on the screen. What they are is a probability claim. On a feed the default is to skip, Lumen's panel puts the population average at roughly two play seconds per attentive second, and that average includes every half-second scroll-past, so the longer a play ran the higher the odds it was attended to.

So a rise in watched rate is a rise in the probability that impressions received active attention, across every impression served. It is not attention measured, and it is not brand lift, which only a lift study reads.

Do we need to buy an attention vendor to get attention data?

Not to manage the account week to week. Stop rate, hold rate and watched rate are computed from metrics already in Ads Manager, three-second plays, ThruPlays or focused views, and impressions, per ad, per platform, per threshold. That is enough to tell which ads earned attention this week and which to retire, which is the decision the data has to support.

Eye-tracking panels and modelled attention scores from vendors like Lumen, Amplified Intelligence or Playground XYZ, all of whom sell the measurement, are the calibration layer, what a second of play is worth in attention on a given platform and format. Useful periodically, not needed daily.

Should we cap frequency?

Cap the right one. Served frequency counts every time the ad entered the screen, including the two-second skips, and most frequency debates are about wearing out an audience most of whom never watched. Watched frequency, the expected number of watched exposures per person, is served frequency multiplied by watched rate, and in most accounts it is a fraction of one.

The lever that raises watched frequency without raising the media bill is the creative, a stronger opening and body lifts the watched rate on the same plan. Meta's own Coca-Cola case, a platform case so read as such, moved ad recall by buying for attentive exposures rather than served ones, and the same logic applies from the creative side.

Sources

• Meta, "Demystifying attention", 2025. Meta's summary of attention measurement methods and standards, the statement on average view time, the Lumen finding that in 75% of cases the highest-attention format did not deliver the highest brand lift (72 ads, 12 advertisers), XPLN's finding on attention time by age (35 campaigns, 2024 to 2025), the Nielsen MMM meta-analysis on executional drivers (119 CPG brands, 2019 to 2021), and the Coca-Cola target frequency case (Meta Brand Lift with Xpln.ai, July to August 2024). Platform sales material, all studies commissioned by Meta, attention figures modelled where stated.
• Havas Media Network, Lumen Research and Brand Metrics, "What Do 9,000 Brand Lift Studies Teach Us About Attention and Memory?", September 2024. Opportunity to see against attention. All three sell attention or lift measurement.
• Lumen Research, cross-platform attention benchmarks presented to the Media Research Group, panel late 2022 to early 2023, as cited in How to read paid social results. The population average of roughly two play seconds per attentive second on Meta and TikTok feeds. Lumen sells attention measurement.
• Dentsu with Kantar and Lumen, attention economy research, as cited in How to read paid social results. A view past two seconds as a choice not to skip, voluntary attention outperforming forced attention. Dentsu and Lumen sell attention measurement.
• Advertising Research Foundation, Attention Measurement Validation Initiative Phase 3, press release, 9 June 2026. Media weight, reach and frequency as conditions for attention to translate into outcomes. Member-funded industry body.
• Media Rating Council, Viewable Ad Impression Measurement Guidelines v1.0, 30 June 2014. The viewable video standard. Industry accreditation body.
• TikTok Ads Manager Help Centre, "About the video views objective". Focused view definition. Platform documentation, TikTok sells the media.
• Amplified Intelligence, "Why does the attention-memory threshold matter?", 27 June 2022. The 2.5-second threshold. Amplified sells attention measurement.
• Byron Sharp, How Brands Grow, 2010. Advertising refreshing existing memory structures for established brands, the reminding job. Ehrenberg-Bass Institute, sponsored by large advertisers.